Bayesian network-driven clustering analysis with feature selection for high-dimensional multi-modal molecular data.

Area of Science:

• Computational biology
• Bioinformatics
• Statistical genomics

Background:

• Rapid accumulation of multi-modal molecular profiling data (genomic, epigenomic, transcriptomic) from bulk tumors and single cells.
• Need for advanced statistical and computational methods to analyze complex, high-dimensional data for biological discovery.

Purpose of the Study:

• Introduce Nebula, a Bayesian integrative clustering method for high-dimensional multi-modal molecular data.
• Develop a computationally efficient variational Bayes approach for model inference.
• Identify directly interpretable clusters and associated biomarkers within a unified framework.

Main Methods:

• Bayesian integrative clustering analysis.
• Variational Bayes approximation for joint posterior distribution inference.
• Application to pan-cancer genomic, epigenomic, and transcriptomic data (approx. 9000 samples).

Main Results:

• Nebula reveals patterns based on shared pathway alterations, offering insights beyond tumor type.
• Demonstrates biological and clinical relevance in a pan-cancer analysis.
• Shows utility in single-cell data for immune cell decomposition.

Conclusions:

• Nebula provides a unified and biologically plausible framework for multi-modal data integration and clustering.
• The method offers advantages over state-of-the-art clustering techniques.
• Nebula facilitates biological discoveries and clinical insights from complex molecular data.